Ask a Human

**Quick judgment** Useful but niche skill: good for low- to medium-stakes *subjective* decisions when you explicitly can tolerate delay and uncertainty. Not suitable for anything that needs an immediate or guaranteed answer. **Key risks & limitations** - **High latency / no guarantee of response**: Answers can take minutes to hours and may never arrive; the agent must always have a fallback strategy. - **Uncontrolled data exposure**: All question content is shown to random humans, so sensitive, identifying, or proprietary data must be carefully redacted or avoided. - **Quality and bias of responses**: Responders are random volunteers; judgments may be noisy, unqualified, or culturally biased. Consensus does not equal correctness. - **Operational complexity**: Requires persistent tracking of `question_id`, polling with backoff, and logic for partial, expired, and late responses. **Recommended scenarios** - When the agent faces a subjective choice (tone, style, preference) and the stakes are modest (UX copy, marketing phrasing, casual communication). - When you want a sanity check or diverse perspective but can continue with a reasonable default while waiting (fire-and-forget pattern). - When a decision is important but can be delayed a few minutes, with explicit user communication about the wait and a clear timeout fallback. **Not recommended for** - Time-critical or safety-/compliance-critical decisions where you need a reliable, timely result. - Any context involving personal, confidential, or legally sensitive data that should not be shared with random third parties. - Situations where consistent, auditable decision-making is required; human crowd preferences may vary over time and across questions.